UPTEC STS15 009 Examensarbete 30 hp Juni 2015

Introduction of Autonomous Vehicles in the Swedish Traffic System

Effects and Changes Due to the New Self-Driving Car Technology

Felicia Bohm Klara Häger Abstract Introduction of Autonomous Vehicles in the Swedish Traffic System Felicia Bohm & Klara Häger

Teknisk- naturvetenskaplig fakultet UTH-enheten Vehicles that are able to drive partly or fully without human interaction are called autonomous. Several Besöksadress: companies work towards an implementation on the Ångströmlaboratoriet Lägerhyddsvägen 1 commercial market. This project studies Hus 4, Plan 0 autonomous vehicles through simulations of road capacity, emissions and fuel consumption together Postadress: with discussions about the implementation in the Box 536 751 21 Uppsala Swedish context. Barriers seen as the most critical are technology, user acceptance and social factors Telefon: and laws and regulations. Simulations of today’s 018 – 471 30 03 conventional vehicle fleet are performed and

Telefax: compared to corresponding simulations with 018 – 471 30 00 autonomous features included. A part of the Uppsala traffic network is simulated and key Hemsida: indicators average delay, average number of stops http://www.teknat.uu.se/student and average speed are studied. Simulation results for a high vehicle flow, corresponding to a maximum hour in the chosen network, show that the implementation will improve the road capacity parameters. Delay and number of stops decrease with 56 respectively 54 percent and speed increases with 34 percent, which are all desirable changes. Corresponding results for a low vehicle flow is a deterioration of delay and speed with 1.3 and 0.38 percent and an improvement of number of stops corresponding to 2.9 percent. Results for the low vehicle flow are not as distinct as for high flow and this pattern repeats in results from calculations for emission and fuel consumption. A workshop is held to discuss autonomous vehicle’s impact on the Swedish urban development. The participants of the workshop contributed with discussions about behavioral changes, conflicts of interest and laws and regulations in terms of autonomous vehicles.

Handledare: Björn Hugosson Ämnesgranskare: Joakim Munkhammar Examinator: Elísabet Andrésdóttir ISSN: 1650-8319, UPTEC STS15 009 Populärvetenskaplig sammanfattning

Världshälsoorganisationen uppskattar att ungefär 1,2 miljoner människor omkommer i trafikolyckor och cirka 50 miljoner skadas i trafiken varje år världen över. Den mänskliga faktorn är involverad i 90-95 procent av dessa olyckor och ungefär 60 procent sker enbart på grund av mänskligt felande. Hur människor agerar i trafiken är därför en central aspekt i termer av trafiksäkerhet och antal olyckor som sker på de svenska vägarna.

Fordon som kan köras delvis eller helt utan mänskligt interaktion kallas autonoma. Detta är en teknologi som utvecklas snabbt, och flertalet företag arbetar för en implementering av autonoma fordon på den kommersiella marknaden inom en snar framtid. Denna rapport syftar att utreda och diskutera hur en implementering av autonoma fordon påverkar det svenska trafiksystemet samt framtida samhällsutveckling. Detta utreds genom att studera vilka barriärer som är kritiska för införandet, hur dessa faktorer påverkar vägkapacitet, bränsleförbrukning och utsläpp samt hur en implementering av autonoma fordon i förlängningen påverkar den svenska samhällsutvecklingen. Rapporten består inledande av en litteraturgenomgång, innehållande artiklar samt seminarier, med syfte att belysa viktiga aspekter gällande autonoma bilar i ett svenskt perspektiv. De aspekter som efter genomförd litteraturgenomgång anses mest kritiska i introduktionsfasen av självkörande fordon är teknologi, acceptans och sociala faktorer samt ändringar i lagar och regler.

Litteraturgenomgången belyser även egenskaper som skiljer de autonoma fordonen från de konventionella. Dessa egenskaper implementeras sedan i en simuleringsmodell där dagens konventionella fordonsflotta jämförs med bilar som har autonoma funktioner. En del av Uppsalas trafiknätverk används för att presentera en realistisk trafiksituation med verkliga vägar, korsningar, trafiksignaler och trafikflöden. Simuleringar av nyckelfaktorerna genomsnittlig fördröjning, genomsnittligt antal stopp och genomsnittlig hastighet utförs i mikrosimuleringsprogrammet Vissim för både ett högt och ett lågt trafikflöde. Resultat för simuleringar med högt trafikflöde visar att en implementering av autonoma fordon förbättrar kapaciteten på vägarna. Fördröjning och antal stopp minskar med 56 respektive 54 procent och hastigheten ökar med 34 procent, vilka alla är önskvärda förändringar för ökad effektivitet. Motsvarande simuleringar för ett lågt trafikflöde ger försämrade värden för parametrarna fördröjning och hastighet med 1,3 respektive 0,38 procent, och en förbättring på 2,9 procent för antal stopp. Resultaten för lågt flöde är inte lika utmärkande som resultaten för högt flöde och i och med att det låga resultaten är av väldigt liten storleksordning kan inga slutsatser dras gällande hur autonoma fordon påverkar kapaciteten på de svenska vägarna vid lågt flöde. Samma mönster upprepar sig i uträkningarna för utsläpp samt bränsleförbrukning där dessa sjunker med autonoma funktioner inkluderade för högt fordonsflöde och är i princip oförändrade för lågt flöde.

Även en kvalitativ del inkluderas i studien, där en workshop genomförs för att vidare diskutera simuleringsresultat samt hur autonoma fordon påverkar den svenska samhällsutvecklingen. Deltagarna i workshopen arbetar alla inom området samhällsutveckling. Diskussionen bidrog med intressanta aspekter som delvis faller innanför ramarna för detta projekt men också aspekter som lyfter fokuset och belyser intressanta följdfrågor gällande hur autonoma fordon i förlängningen påverkar samhället. Beteendemönster, intressekonflikter samt lagar och regler är de aspekter som återkom frekvent under diskussionen och därmed anses vara de mest relevanta.

Table of Contents

1. Introduction ...... 1 1.1 Aim and Objectives...... 2 1.2 Disposition ...... 2 2. Background ...... 3 2.1 Drive Me...... 3 2.1.1 Levels of Automation ...... 3 3. Methodology...... 5 3.1 Literature Review...... 5 3.2 Simulations ...... 6 3.2.1 Vissim ...... 6 3.2.2 Implementation of Traffic Environment ...... 6 3.2.3 Simulation Scenarios ...... 8 3.2.4 EnViVer...... 8 3.3 Workshop...... 9 3.4 Limitations and Credibility...... 9 4. Underlying Concepts...... 11 4.1 Man, Technology, Organisation - MTO...... 11 4.2 Value of Travel Time...... 11 5. Autonomous Driving ...... 12 5.1 Technology and System ...... 12 5.2 Adoption and Time Horizon ...... 14 5.3 Implementation Effects ...... 15 5.4 Environmental Aspects ...... 16 5.5 Electrical Cars...... 17 5.6 Laws and Regulations...... 17 5.7 New Models ...... 18 5.8 Opinions from Transportforum 2015...... 19 6. Simulation Data...... 21 6.1 Vissim Data...... 21 6.2 EnViVer Data ...... 24 7. Results ...... 26 7.1 Vissim High Flow ...... 26 7.2 Vissim Low Flow ...... 30 7.3 EnViVer Emissions ...... 32 7.3.1 Fuel Consumption...... 33

8. Discussion...... 34 8.1 Adoption and Acceptance...... 34 8.2 New Conditions...... 35 8.3 Workshop Discussion ...... 37 8.4 Future Research ...... 38 9. Conclusion ...... 39 References...... 40 Appendix I, Vehicle Flows in Uppsala ...... 43 Appendix II, Signal Plans & Signal Switching Charts...... 44

Nomenclature

AV - autonomous vehicle CV - conventional vehicle VMT - vehicle miles travelled V2V - vehicle to vehicle communication V2I - vehicle to infrastructure communication

1. Introduction

The World Health Organisation estimates that about 1.2 million people worldwide are killed in road crashes and another 50 million are injured in traffic every year (World Health Organisation, 2013). About 21,000 people died or were seriously injured during 2013 in car accidents in Sweden (Transportstyrelsen, 2013). According to previous research the human factor is involved in 90-95 percent of all accidents and about 60 percent are caused entirely by human error (Forward, 2008). Human interaction is therefore a central aspect in terms of increased road safety and reduced crash rate.

Another central aspect in terms of an improved driving environment and a more efficient transport system is user convenience. Imagine having your daily travel time free for other tasks than driving, such as finish work tasks, take part in important online meetings, sleep, or watch a movie. The car will safely navigate itself to the chosen destination and you do not even have to park the car when you reach the destination (KPMG, 2012).

Both high technological companies, such as Google, and more traditional automotive manufacturers, such as Volvo, are developing and testing technologies for vehicles that are able to drive partly or totally without human interaction. These vehicles are called autonomous and this is a fast developing area where new technological breakthroughs are announced more often than expected by experts (KPMG, 2013). The technology is not commercially implemented in the Swedish traffic system but there are companies planning to introduce test fleets in a near future. One example is Volvo’s pilot project “Drive Me” which aims to implement a fleet with 100 autonomous vehicles in the city of Gothenburg by 2017 (Volvo Car Group, 2013).

Increased safety and convenience are two of the most revolutionising benefits that humans will gain from the introduction of self-driving cars. These cars are also considered to reduce congestion, increase the efficiency of land use and increase road capacity (Rigole, 2014). These great benefits are the ones most frequently communicated to the public, but other important aspects regarding autonomous vehicles will also strongly affect society either in a positive or negative way. Fuel consumption and emissions may change as driving behaviour changes. All these aspects need to be considered in the analysis about how the autonomous vehicles will affect society.

1

1.1 Aim and Objectives

The aim of the thesis is to analyse how an implementation of autonomous vehicles in the Swedish context could affect the urban development and the existing traffic situation. The study contains qualitative aspects as well as simulations of essential key indicators regarding roadway capacity, emissions and fuel consumption. The aim is to highlight changes caused by the implementation by answering the following questions:

. What barriers are the most critical when introducing autonomous vehicles in the Swedish traffic system? . How does an introduction of autonomous vehicles affect road capacity, fuel consumption and emission rate of the Swedish car fleet? . What aspects regarding the implementation of autonomous vehicles will influence the urban planning and development in Sweden?

1.2 Disposition

The report begins with a background that aims to give an introduction to autonomous vehicles. The next section presents the method used in the project containing literature review, simulations, workshop and limitations and credibility. This is followed by underlying concepts, which present the theoretical concepts used for the analysis. Next section continues with autonomous driving, which consists of aspects and opinions regarding autonomous driving carried out through the literature review. Simulation data comes next and presents explanations and numerical values for all parameters used for simulation and emission calculations. This is followed by presentation of simulation results. A discussion part containing all previous sections and discussion from the workshop concludes the report before references and appendixes.

2

2. Background

Autonomous driving generally refers to vehicles with the ability to control the driving without human interaction. This gives humans the possibility to pass on driving tasks or the complete driving to the car’s control systems, but can also include the vehicle taking over the control in situations where the driver cannot cope (Nåbo et al., 2013). Technology giants and automobile manufacturers are currently working toward complete automation where the aim is to sell cars that can drive safely and effectively to a destination in respect to current roadway conditions without any human interaction (Pinjari et al., 2013).

2.1 Drive Me

Volvo has, together with other companies, governments and research groups, initiated the project Drive Me, which aims to create mobility in a sustainable society. The level of automation in the project is “highly autonomous cars”, which can be compared to level 3 (explained in section 2.1.1). This means that the driver must not monitor the vehicle at all times but is required to cede control if needed in short notice. The system enables the vehicle, without a person driving it, to find a parking space and park itself. A test fleet of cars at this level will enter the roads of Gothenburg in 2017. Drive me includes all key factors regarding autonomous driving; legislation, transport authorities, a city environment, a vehicle manufacturer and real customers. The test fleet will act in everyday driving conditions on 50 kilometres of selected roads in Gothenburg typical for commuting, including both freeway conditions and frequent queues. Volvo highlights that these everyday driving conditions will provide insight into the societal benefits that are central for the company to be able to offer sustainable personal mobility in the future (Volvo Car Group, 2013). To avoid queues and to get a smoother traffic flow the cars will be connected to the traffic management centre of Gothenburg. New features in these cars are adaptive cruise control and steering assistance which automatically will keep up with traffic without the driver’s intervention. The driver is able to decide whether he or she wants to drive manually or use automated driving (Automotor & Sport, 2013).

2.1.1 Levels of Automation

The US agency National Highway Traffic Safety Administration (NHTSA) has defined five levels of automation of vehicles (NHTSA, 2013). The following section describes each level and contains an increasing amount of automation.

Level 0 - No automation Level 0 means that the driver is in complete control of the vehicle, there is no automation. The driver is fully responsible that the vehicle is driven safely. Some

3 devices may be included, for example blind spot monitoring and windshield wiper, but the vehicle is unable to self-control steering, brake or gas (NHTSA, 2013).

Level 1 - Specific function automation Level 1 indicates moderate automation and helps the driver with certain specific functions, such as automatic braking or lane positioning. The vehicle can have more than one automated function but the functions are not combined, they work independently. The driver is still fully responsible for the safety of the driving (NHTSA, 2013).

Level 2 - Combined function automation Combined function automation defines level 2, and implies that the vehicle has the ability to address at least two specific functions in combination to assist the driver, for example lane position and cruise control. Thus, drivers who drive long distance may benefit at this level. The driver must still maintain full supervision of the road and other vehicles and is expected to be alert and respond at short notice (NHTSA, 2013).

Level 3 - Limited self-driving automation Level 3 is self-driving automation that is limited, which indicates that the driver does not need to constantly monitor the road and can cede control if needed. Accordingly, the driver need to be alert and capable to take over the driving if needed. At level 3 there is an obvious change for the driver, who can direct his or her attention elsewhere for longer periods of time. The system can tell the driver, in appropriate amount of foresight, that he or she needs to safely take control over the vehicle if the automation system requires. An example may be a construction area where the system has difficulties supporting the automation (NHTSA, 2013).

Level 4 - Full self-driving automation Full self-driving automation corresponds to level 4. The car is structured so that all driving functions are monitored by the vehicle and only require the driver to set the destination. This level will improve driver experience notably at several aspects. One change is that level 4 vehicles may not look like the conventional vehicles. All persons in the car are travelling as passengers, and will therefore be able to focus the attention elsewhere. Responsibility for the safety now rests with the vehicle (NHTSA, 2013).

4

3. Methodology

Three different methods have been used in order to answer the research questions stated in section 1.1. The first question was further analysed through a literature review with the aim to carry out basic knowledge of the subject and served as a preparation for the following subprojects. Central aspects regarding AV’s were enlightened and these aspects are further discussed in the end of the report with basis in the underlying concepts presented in section 4. After completing the literature review relevant key performance indicators and questions were formulated based on knowledge from the readings.

Road capacity, fuel consumption and emission rate for autonomous vehicles are determined through simulations of selected key indicators. The simulations were performed in the software Vissim by PTV. The simulation process started with a base scenario simulation corresponding to today´s Swedish vehicle fleet. Autonomous features carried out from the literature review were then implemented in the Vissim model and the results were continuously compared to values from the base scenario simulation. Emission calculations were performed using the software EnViVer which determines emissions of carbon dioxide, nitrogen oxide and particle pollution. These values were then used to calculate fuel consumption of the vehicle fleet.

The project also consisted of a qualitative part in order to discuss question number three. This consisted of a workshop with experts in the urban development area. The purpose was to broaden the scope of autonomous vehicles. Simulation results were presented and discussed and central aspects regarding autonomous vehicles were enlightened in order to determine how the introduction will affect the Swedish urban development.

3.1 Literature Review

New technologies come with both advantages and disadvantages and the information, which was obtained varies from actor to actor. Both commercial and scientific articles and the traffic conference Transportforum have been studied in order to enlighten different aspects and opinions regarding the introduction of AV’s. Articles in the area contain speculations and opinions because of the lack of true data and information regarding the new technology. The literature review therefore presents several perspectives regarding autonomous vehicles. Both technological and social aspects have been taken into consideration while going through the material resulting in a wide range of aspects that were implemented in the simulation model as well as further discussed and presented in section 8. A large part of the thesis work was to carry out how autonomous vehicles will work and how to implement these features into the Vissim micro simulation model. The literature review considered articles and discussions from Transportforum. These sources together with studies of the software parameters resulted

5 in a number of parameters that could be changed to imitate driving behaviour of autonomous vehicles in Vissim.

3.2 Simulations

Simulations of a traffic network were performed in order to determine how an introduction of AV’s can affect the traffic environment and the build environment. The studied key indicators were average delay, average number of stops and average speed. Simulations of these indicators were performed in the traffic simulation software Vissim by PTV and the results were then used in the software EnViVer to calculate resulting emissions.

3.2.1 Vissim

The software Vissim is used to simulate private transport, goods transport, public transport, pedestrians, cyclists and their interactions. It is possible to set driving characteristics for vehicles which makes the model suitable for complex traffic situations. The software performs microscopic simulations and displays the results in reports as well as graphic presentation through 2D and 3D visualisations (PTV Group, n.d). Vissim has previously not been used for simulation of autonomous driving with the same purpose as for this project. The software is detailed and several parameters can be changed in order to implement specific driving behaviour and it is therefore considered as suitable to mimic autonomous features and consequently to fulfil the purpose of the thesis.

Vissim has two built in car-following models, Wiedemann 74’ and 99’, which explain vehicle driving behaviour. The models are similar to each other besides that the 99’ model is more adjustable. To describe the differences in car-following the models use thresholds. Some of these thresholds use the speed difference between two vehicles to analyse driving behaviour, since the thresholds vary over different speeds (Higgs et al., 2011). Vehicles are in this project represented with Wiedemann 99’ due to more adjustable parameters that describe the automation of vehicles in more detail.

3.2.2 Implementation of Traffic Environment

The aim of the simulations was to gather data about changes that will occur when introducing autonomous vehicles into the existing traffic situation. An important aspect was therefore to model the traffic system as realistic as possible to determine changes that will be reality with the introduction. The simulated traffic network is based on parts of a real traffic situation in the city of Uppsala in Sweden. Uppsala traffic situation was chosen because the authors are well aware of this area and were therefore able to reproduce a simplified network that is close to the real situation in the area. The basic idea when choosing the network was to include both complex traffic junctions as well as higher speed areas and that these links should be representative for a real traffic network. Simplifications of the original network were made and busy roads were

6 considered the most relevant due to clearer results in situations with higher vehicle flows. Two different vehicle flows were simulated in order to determine changes connected to the introduction for both high and low traffic flows. The vehicle flows were modelled with real flows in the chosen network considered. Data collected and provided by Uppsala Municipality contained hourly numbers of cars passing certain points in the network (see Appendix I). Vehicle flows used in the model correspond to the real values to make sure that they were of the same order as for the real traffic system. The freeway is owned by the Swedish Transport Department and data for vehicle flows on this section is therefore provided by them (Trafikverket, 2013b). Both a high flow and a low flow scenario were considered in the simulation process. High flow corresponds to the highest vehicle flow on several measurement points in the network. Data was provided for one week and the highest hourly flow is the one used for the first simulation. The low flow scenario is based on 30 percent of the number of vehicles in the high flow, which was determined by the average percentage difference between a high flow and a low flow hour. The lowest hourly number of cars passing the measurement point was close to zero and the 30 percent approximation was therefore implemented in order to determine results for both high and low vehicle flow.

Despite simplifications, the system and traffic flows are considered as realistic and representative to fulfil the aim of the report. The simulated network contains five intersections with different characteristics, a roundabout, and a highway stretching 6 kilometres (see figure 1).

Figure 1. The simulation network implemented in Vissim. A simplified part of Uppsala traffic system.

7

To further keep the network close to a real traffic situation signal controllers were set according to the actual traffic signals in the chosen intersections. Signal plans and signal switching charts provided by Uppsala Municipality were studied and implemented in the Vissim simulation (see example of an intersection in figure 2). Examples of signal plans and signal switching charts are presented in Appendix II.

Figure 2. Intersection based on the real traffic situation with traffic signals included in the simulation.

3.2.3 Simulation Scenarios

The simulations in Vissim started with the base scenario where settings represented a conventional vehicle fleet in the Swedish context. Parameters specific for autonomous cars were then changed one at the time resulting in a number of simulations with different features enlightened. The aim of modelling one parameter at the time was to enlighten how much the different features affect the result, before putting all autonomous features together.

After performing simulations with one parameter change at the time the results were analysed and the parameters were considered together. This resulted in further simulations containing a number of features that correspond to a realistic scenario where autonomous vehicles are introduced in the current traffic network. Explanation and numerical values of the changed parameters corresponding to autonomous features are presented in section 6.

3.2.4 EnViVer

EnViVer is a software developed to calculate real world vehicle emissions. Data from a Vissim simulation can be imported into EnViVer. The data fields that are required from Vissim and used in the emission calculations are number of vehicles, number of vehicle type, name of the vehicle type, simulation time, speed and position. The emission measurements are updated continuously to the latest vehicle technology. Air condition, cold started engines and age are some of the factors that are analysed and taken into

8 account when calculating the emissions. EnViVer is based on Versit+ which is a road traffic emission model developed by TNO. It calculates the emissions of carbon dioxide, nitrogen oxide and particle pollution (CO2, NOx and PM10) based on instantaneous velocity and acceleration. This makes Versit+ the perfect tool to evaluate the effect of traffic management measures (TNO, n.d). Parameters that are changed in the EnViVer emission model are presented in the section 6.2.

Calculated emission results were further used to calculate fuel consumption for the base scenario and the scenario with all parameters changed. Data for average carbon dioxide emissions per fuel type in Sweden were used to calculate fuel consumption per kilometre travelled.

3.3 Workshop

A workshop with experts in the urban development area was held in order to broaden the scope and discuss how the introduction of autonomous vehicles will affect society. The participants in the workshop were in addition to the authors Björn Hugosson, consultant in vehicle and traffic strategy, Patrik Tornberg, consultant in urban planning, and Maria Nilsson, consultant in traffic and energy systems. All participants had beforehand received the results of the simulations and the workshop also started with a review of the results with the aim to give all participants the possibility to ask questions. An open discussion method was considered as the most favourable because of the low number of participants. All opinions were heard during the discussion and the participants all contributed with important inputs. Discussion in group gives the participants the chance to interact and inspire each other, which generates new perspectives. An open discussion therefore implies expanded thoughts and a more detailed discussion (Jacobsen, 2002).

3.4 Limitations and Credibility

Since fully automation of vehicles is a relatively new field, the report contains several assumptions. The simulations are based on a basic current traffic network that is simplified to fit within this project. All details regarding an already existing traffic situation are therefore not considered. No other entities than cars, buses and trucks are included in the simulations and for example pedestrians and bikes are not taken into consideration in this project. Some aspects that are planned to be included in autonomous vehicles that will enter roads in the future are not implemented in the simulations. One of these aspects is communication between vehicles and the surrounding traffic network, which is a feature that implies that the driving will be further planned and coordinated concerning a high penetration rate. A possible result of a scenario where vehicles can communicate with each other and infrastructure is that traffic lights and such can be removed. Because of the absence of communication in the simulations the traffic signals are considered as unchanged from today’s conditions. Another aspect that also relies on communication to maintain security is platooning

9

(explained in section 5), which is therefore not included in the simulations. The autonomous features that are not included in the simulations are logically not presented in the result data, although this is discussed later in the report.

Penetration rate of AV’s is another aspect that is not considered fully in the simulations due to software limitations. To enlighten the consequences of an introduction of autonomous vehicles a penetration rate of either zero percent in the base case and 100 percent in the rest of the simulations is considered. A smaller proportion of fully automated vehicles would be more realistic reflecting about the closest future. The effects regarding AV’s become clearer in a scenario with a higher adoption rate which makes it reasonable to consider cars either conventional or autonomous in the simulations.

The study contains parameter values that have been estimated based on knowledge about autonomous vehicles and possible parameter adjustments in the software. There are no experiments stating exact numerical values of parameters implemented in Vissim, due to the new technology regarding AV’s and the originality of the simulations in Vissim. All parameter values are though estimated based on realistic scenarios and experiments which imply that the result of the study is accurate and corresponds to the result that would follow from an implementation. Even though assumptions are made the result is realistic, hence it is necessary to have these assumptions in mind when considering the overall results. Resulting values do not numerically represent the exact numbers of the key indicators that will follow from an implementation of AV’s. Ratios and proportions between the conventional and the autonomous fleet are more accurate to consider than absolute numbers due to that the same basic conditions have been used for the fleets. Results are both presented in absolute numbers and proportions to give a clear understanding of the simulation results and calculations.

10

4. Underlying Concepts

This chapter presents the underlying theoretical concepts that are important to have in mind when going through the material carried out from the literature review. The first concept consists of aspects regarding safety and responsibility and the second part is about travel convenience.

4.1 Man, Technology, Organisation - MTO

A commonly used method when performing accident investigations is the MTO- method. This method is usually applied to investigate the underlying causes to accidents that have occurred. In this report the MTO-theory is used proactively with the purpose to determine who would be responsible in case of an accident regarding autonomous vehicles.

The base of MTO-theory is that all three parts, Man, Technology and Organisation, shall be borne equally in the analysis. All these elements are crucial for the events leading to the accident and they should all be taken into consideration. There is usually a single human or a group of humans who get the blame for an accident but the underlying technological and organisational factors can be as important. The first thing to do in an MTO-analysis is develop the event sequence of the accident, and then identify human and technical causes of these events. If an accident occurs it is important to investigate what and who caused it, in terms of human, technology and organisation, to prevent a recurrence (Sklet, 2004).

4.2 Value of Travel Time

Automated transportation that results in more time available for the driver to engage in other tasks than driving also implies a change in value of travel time. This term refers to the cost of time spent on transport and includes unpaid personal travel time as well as costs to businesses of paid employee time spent on travel. Both waiting time and actual travel time are included and the value varies depending on trip purpose, travel conditions and user preferences. One of the factors that affect travel time value in a positive manner is when public transit travel time is enjoyable and productive. Time spent in discomfort implies a higher cost per minute than travel time spent in comfort, and passengers tend to work, read or rest under such pleasant conditions. The value of time varies for different people depending on preferences and needs. Some people appreciate driving while others place a higher cost on conventional driving than on public transport. Previous research has shown that people seem to enjoy about 30 minutes of travel a day even though average travel time is about 60-90 minutes and people dislike devoting more than 90 minutes a day on travel (Litman, 2009).

11

5. Autonomous Driving

Several automated systems are used in conventional vehicles today, such as adaptive cruise control, autonomous emergency braking and parking assist. The current sensing technology in automated vehicles is sensitive, and do not work perfect, in complex traffic situations and severe weather conditions and need to be more reliable to make a fully implementation possible (International Transport Forum, 2014). Laws and regulations need to be changed before full self-driving vehicles, level 4, can become reality on Swedish roads. Changes in regulations and technology, together with further development is needed before fully automated vehicles can enter the commercial market. This section presents information about the implementation of autonomous vehicles in the Swedish context collected during the literature review.

5.1 Technology and System

Technologies for self-driving cars can differ slightly depending on manufacturer and system choices, but all systems have the same purpose. They all aim to sense the environment around the vehicle (Ibañez-Guzman et al., 2012). Self-driving vehicles use the same technology that is used in many robotic systems, a so called “sense-plan-act” design (RAND, 2014). To determine range to obstacles and find a drivable path to the given destination the operating system in an autonomous vehicle uses LiDAR (Light Detection and Ranging), radar, cameras, and GPS sensors (Rigole, 2014). These appliances gather data about the environment and the car’s software interpret the data to plan the actions of the vehicle. Commands are then sent to the vehicle’s control system that navigates steering, acceleration and brakes (RAND, 2014). LiDAR sensors measure distance to vehicles or other obstacles by illuminating them and analysing the reflected light (KPMG, 2012). This tool can measure speed, distance, rotation and chemical composition with corresponding concentration (Bloomberg Business, 2015).

Features that will be included in self-driving cars are vehicle to vehicle (V2V) communication and vehicle to infrastructure (V2I) communication. The V2V and V2I systems contain wireless communication links that make it possible for vehicles to interact with each other and the network to plan the journey in more detail. The possibility to communicate with others extends the vehicle’s awareness of the environment and the detection is therefore more detailed than detection that is only sensor based. Vehicles using V2V and V2I communication will react in advance of obstacles, both with respect to other entities and to changes in the external circumstances (Ibañez-Guzman et al., 2012). An improved V2I system will provide the vehicle with information about for example the traffic situation ahead, road weather, traffic signals, infrastructure asset information, such as pavement markings, and impending congestion. This information can then be used to make better route choices and disperse vehicles more efficiently through the network (Peters, 2013).

12

Ibañez-Guzman et al. (2010) divide vehicle navigation into four basic functions (see figure 3) that need to operate properly for the vehicle to navigate safely and in an optional manner to the planned destination. The base function localisation determines the vehicle's position and orientation with respect to either an absolute or a relative reference coordinate frame. Navigation systems, such as GPS, provide absolute positioning information and relative information can be expressed in relation to other vehicles or position in a traffic junction. Mapping is a base function that aims to understand the relationship between the vehicle and the environment in terms of space and time. The function can be compared with the human driver’s ability to perceive the environment which is to be travelled and create a mental model based on knowledge from previous situations and subconscious recalls to traffic legislation. The understanding gathered is then used to determine where the vehicle can move. The base function that enables the vehicle to move safely and efficiently to a given destination is called motion. The function includes path planning and avoidance of obstacles, which result in determining the vehicle trajectory, and the vehicle control, which aims to move the vehicle to a destination. The last base function is interaction and contains the vehicles ability to move in a network shared with other entities such as pedestrians, other vehicles and bicycles. The underlying objective in this function is to avoid collisions with other entities, reduce the driver’s anxiety and ensure a safe and efficient traffic flow. None of these basic navigation functions can perform poorly in an autonomous vehicle. All cars and other entities are expected to obey a set of traffic rules regardless of whether the vehicle is automated or not, which set high demands on the AV’s control system (Ibañez-Guzman et al., 2010).

Localisation Motion

VEHICLE NAVIGATION

Mapping Interaction

Figure 3. Ibañes-Guzman et al. (2010) have divided vehicle navigation into four fundamental navigation functions.

13

Even though successful test runs have been performed, for instance Google has logged more than 800,000 kilometres of autonomous driving without crashes, the technology for completely autonomous vehicles is still facing some difficulties that need to be solved before the driver can be totally uninvolved in driving tasks. The sensor system used can gather a lot more information than a human, but the issue when using a sensor system is to turn the gathered data into clear understanding of the environment. Another critical aspect in terms of autonomous vehicle functions is that all required sensors work properly in use and that failure of these are detected. An autonomous vehicle depend strongly on its sensors and failure due to electrical issues, physical damage or ageing components need to be found by inbuilt sensing algorithms that can determine inadequate performance (RAND, 2014).

The proposed pricing for an autonomous vehicle seems to differ depending on the actor presenting the information. According to Volvo the price surcharge for level 3 AV’s will be between 20,000-40,000 SEK (Automotor & Sport, 2014). The LiDAR system used in the Google car is said to cost 70,000 USD (approximately 590,000 SEK) (KPMG, 2012). System choices and level of automation will influence the pricing of the vehicles, and the technology will probably be more expensive in the beginning and decrease when the adoption rate increases (Bierstedt et al., 2014).

5.2 Adoption and Time Horizon

A certain adoption rate of AV’s is needed to increase road capacity. Researchers have brought up a variety of proportions, but they all agree that a moderate extent of AV’s is necessary for a notable increased capacity on the roads. According to Pinjari et al. (2013) the capacity is increased by 22 percent at 50 percent market penetration, and with 100 percent AV’s the capacity is estimated to increase by maximum 80 percent. A mass penetration of fully automated vehicles is not possible for at least a few decades. There are some factors that need to be considered to make it possible (Pinjari et al., 2013). Consumers must adopt the technology and allow the vehicle to take over control, and it can be expected that a proportion of the population is reluctant to use driverless cars. User acceptance must therefore be increased since it affects the impact of the technology (Kungliga Tekniska Högskolan, 2014). At the traffic conference Transportforum 2015 the policeman Bengt Svensson expressed the risk that people driving CV’s (conventional vehicles) will intrude and drive in front of AV’s as self- driving cars always give way to cars nearby as they must have a certain safety distance to the surrounding environment, which might impair today’s efficiency (Transportforum, 2015). AV’s communicate with each other in order to streamline the driving which require that a certain proportion of the road traffic is self-driving. One example is on-ramps to highways that often form bottlenecks, if vehicles could communicate at weaving points the capacity may be increased (Lind et al., 2014).

Experts in the area have different opinions and thoughts about the introduction of autonomous cars. Several companies, both automobile manufacturers and technology

14 companies, predict that they will release the technology by 2016 and sell fully autonomous cars in 2020. Experts from the Institute of Electrical and Electronics Engineers (IEEE) express one of the most common predictions, that 75 percent of all vehicles will be fully automated by 2040. However, Todd Litman of Victoria Transport Policy Institute believes that it will take until 2060 to reach 75 percent of the vehicles are autonomous. At what decade the market consists of a large proportion of autonomous vehicles depends partly on technological aspects due to that more testing and development is needed before AV’s are ready for public use. New technologies tend to be expensive and only affordable for people with a high income, which implies that the economical aspect also is central for implementation time. Laws, infrastructure changes, personal preferences by humans and computer security will affect the adoption rate and will therefore strongly affect the time horizon for a large proportion of autonomous vehicles on the roads (Bierstedt et al., 2014).

One downside people fear is the risk of intrusion into the data systems and thereby the risk that the network will be brought down. Consumers are worried that intruders could track their vehicles, give them false information or endanger their private life, among many other security threats. This could be prevented by, for example, delete data that could identify the driver, summarise data within the vehicle rather than sending the information further or user-defined integrity policy. There are also some advantages with data collection, such as analysing patterns in road use and driving behaviour in order to make enhancements (KPMG, 2012).

5.3 Implementation Effects

The human factor is, as mentioned earlier, involved in 90-95 percent of all car accidents. Autonomous vehicles will likely reduce the number of crashes due to the sensor system that will be able to control the vehicle with more precision than a human driver (Lind et al., 2014). Reduced crash rate might further lead to lighter cars because vehicles tend to be heavy in order to prevent injuries in case of a crash. Lighter vehicles imply reduced fuel use, improved fuel economy and reduced emissions. Fuel cost can also be reduced by more smooth acceleration and deceleration when the vehicle is able to plan the driving with respect to other vehicles and obstacles (Anderson et al., 2014). However, there are researchers who believe that vehicles will become larger when they become autonomous, and this is because people can engage in other activities than driving. Pinjari et al. (2013) think that more space will be needed inside the vehicles for things like televisions and computers, and some vehicles may include WC and sleeping area. Due to the possibility of engaging in other activities while travelling people may be more willing to travel farther, as they can use the travel time efficiently (Pinjari et al., 2013). There is research stating that this might lead to increased vehicle miles travelled (VMT) and thereby increased fuel consumption, congestion, emissions and suburban sprawl (Anderson et al., 2014). Another effect of AV’s, and that people may travel farther due to more effective and pleasured travel time, is that companies could place

15 their business farther away from the city centre where it is less expensive to locate activities, which will also increase the VMT (Pinjari et al., 2013).

Implementation of autonomous cars could change the structure of the current roads and land use. By enabling automated parking the vehicle can drop passengers off and locate an available parking space and position itself close to another vehicle, thus a smaller parking area would be needed. This could also reduce the need of car parks in highly valued areas, such as within the city limits (Anderson et al., 2014). The traffic lanes may also be changed, due to lane keeping support the lanes can be narrower. Initially autonomous vehicles might require separate lanes where they would not be mixed with today’s CV’s (Lind et al., 2014). A large penetration rate of AV’s is though needed for vast infrastructure changes to be beneficial (Pinjari et al., 2013).

In the long term autonomous vehicles are assumed to be completely self-driving without human interaction. If this becomes reality new opportunities open up for those who are currently unable to drive. Elderly, young and disabled might be able to travel individually which improves the independence. Another effect with a large proportion of autonomous vehicles is that the distances between vehicles can be reduced and therefore result in an increased roadway capacity (Anderson et al., 2014). This implies that in a future scenario cars will be able to communicate with each other and therefore plan the driving with respect to what is happening ahead (Michael et al., 1998). Platooning is a feature where vehicles drive close to each other in high speed which reduces the aerodynamic drag for the following vehicles, which implies overall lower fuel consumption (Brown et al., 2014). California PATH demonstrated as early as 1998 that the safety distance between vehicles driving in a platoon increases as the communication delay between the vehicles increases. This states that a shorter distance between vehicles is possible with constant, or increased, safety when communication is included (Michael et al., 1998). Platoon driving has been tested in California, with speed up to 105 km/h the distance between cars was set to 6.5 metres, which is equal to a 0.2 second time gap (Lind et al., 2014). Experiments that have been performed using trucks show that a decreased distance between vehicles to four metres implies energy savings by 10-15 percent compared to average distances in today’s driving. Safe platooning that results in this decrease will require a high performance sensor system and communication such as V2V and V2I (Barth et al., 2014).

5.4 Environmental Aspects

Whether the energy consumption will be affected positively or negatively by the implementation of AV’s depends on several aspects. If the driving is optimised, for example through cruise control or smooth acceleration and deceleration, the fuel usage can be more efficient and the emissions reduced (Anderson et al., 2014). Optimised driving is also called eco-driving and recent analyses identify potential fuel savings of 20-30 percent for the most aggressive drivers. Savings would be less for drivers that already apply eco-driving but AV’s will be able to constantly drive efficiently and

16 smooth which most likely will improve the overall fuel use additionally. Communication features such as V2V and V2I provide the ability to plan driving more carefully and apply smart routing, which could reduce the stop time and make the traffic flow more efficient. Energy usage will therefore be reduced due to smart route choices like avoidance of traffic, use of slower but shorter routes and use of routes with fewer stops. A vehicle fleet containing only fully automated vehicles will also infer smart intersection control. Traffic lights can be removed and vehicles guided through intersections by communication without stopping (Brown et al., 2014). If automation of vehicles can make cars smaller and lighter, experts estimate that the fuel consumption can be reduced by twice as much as for CV’s. A changed driving behaviour regarding AV’s will though affect the total VMT which will affect fuel consumption and amount of emissions. Whether the number of VMT will increase or decrease will be further discussed in section 8. In a scenario where the VMT is increased the fuel consumption will increase. But if the driving is optimised and if the cars are driven with more environmentally friendly fuels, even the impact of increased VMT can have neutral effects on the environment (Anderson et al., 2014).

5.5 Electrical Cars

AV technology might enable alternative vehicles and fuels, such as electric cars. The battery used to store electricity in today’s cars is heavy but due to reduced weight AV’s may help the transition to electric vehicles. A lighter car would lead to less power needed to move the vehicle forward, which in turn would enable the batteries to be smaller, lighter and cheaper. At a level 4 automation vehicles could require a much smaller battery to travel the same distance as a vehicle of today’s size (Anderson et al., 2014).

Tesla Motors has developed an electric vehicle with automated systems such as self- parking and lane-keeping. The car has cameras and radars to adapt to surrounding traffic and to detect obstacles. If an accident is about to happen the driver will get alerted by an alarm and the computer system will try to swerve (Pasztor, 2014). The car has a stop- and-go adaptive cruise control which means that it follows the speed of the car in front, if one car slows down the following cars slows down (Anthony, 2014). One advantage with automated electric vehicles is that the car itself could recognise and plan when it is time to recharge, the car would also know the locations and accessibility of charging options (Brown et al., 2014).

5.6 Laws and Regulations

It is not possible to introduce completely self-driving cars with respect to today’s laws and traffic rules. The European Union is responsible for establishing fellowship rules, and the United Nation Economic Commission for Europe (UNECE) establishes the technical requirements for vehicles. At the national level the Swedish Transport Agency, the Transport Department and local authorities develop the design of future

17 infrastructure. These actors study the social benefits of autonomous driving in terms of performance, robustness, urban development, environment and health, usability and safety. Stakeholders in Sweden highlight that the road safety will increase when reducing human errors by introduction of autonomous vehicles. What requirements, regulations and other policy instruments that need to be changed are essential aspects to make an implementation possible, and the Transport Agency need to continue to increase their knowledge and participation in the matter (Transportstyrelsen, 2014). There are laws saying that someone must be responsible for the safety of the vehicle. The Road Traffic Convention (also known as Vienna Convention on Road Traffic) demands that all vehicles should have a driver, and that the driver at all times should be able to take control over the vehicle. The regulations also state that vehicles must not be driven by a person who, because of illness, the influence of alcohol or other drugs, exhaustion or other reasons, cannot drive a vehicle safely. If an accident occurs, due to negligence or by intention, it is the driver who is responsible and brought to justice (Transportstyrelsen, 2014). Autonomous cars still require a human’s observation, and the difficulty of getting full attention back from a distracted driver is an issue when the driver has the opportunity to relax and not be aware of the situation on the road, which might result in decreased road safety (Knight, 2013).

When it comes to safety and responsibility there are disagreements whether automation of vehicles has a positive outcome or not. One way to balance the responsibility is to support the driver instead of letting the vehicle take over completely, so called adaptive automation. This adaptive automation is used to assist the driver if the safety requirements are exceeded. According to researchers who tested different levels of adaptive and automated driving, the best results were obtained when the automated system let the driver be involved in the driving. Through shared control over the vehicle the driver do not have to re-engage if the system should fail, which could help eliminate accidents. If the technology would get more responsibility over the driving there are also researchers who believe that it could get a positive effect for some people. For example drivers who drive hazardous due to tiredness or lack of skills (National Highway Traffic Safety Administration (NHTSA), 2013).

5.7 New Models

A possible future scenario is that cars can drive themselves without human presence. This scenario would make new models for car ownership possible and expand the opportunities for vehicle sharing. A vehicle can be summoned when needed and when the destination is reached the car is returned to other obligations (KPMG, 2012). Google recently announced their idea about driverless cars patrolling through residential areas to pick up and drop off occupants. The passenger calls a vehicle and communicate desired trip and the car will get the passenger to the preferred destination. Compared to owning a car this would be a less expensive alternative (Bloomberg Business, 2015). Car sharing is one way to make travel more efficient, it can reduce congestion, lower the amount of fuel used and the transport costs (Rigole, 2014). The Swedish National

18

Road and Transport Research Institute (VTI) have compared how land use differs for different modes of transport. Number of occupied road square meters per road user is for a bus 2.1 m2, and for a car 22.1 m2. Thus, with respect to land use, it is about 10 times more efficient to take the bus than to drive a car (Johansson, 2004).

In the Netherlands an automated transport system is used that works without any physical guideline, called The ParkShuttle. The vehicles use electrical navigation instead of computers and to lower the costs they travel along a simple 3 meter wide asphalt lane. The vehicle has a sensor system which will stop the vehicle if an obstacle occurs, and the speed limit is set to 30 km/h. This mode of transport is not suitable for longer trips, it works best for shorter distances, 500 m - 5 km, such as between the workplace and a public transport station. There is room for 10 passengers and the vehicle will not depart if the maximum number of passengers is exceeded. A pilot program is currently in progress in Rotterdam where a 1,200 meter long link between a business park and a metro station is under construction. Goals with the project are, among other aspects, to make public transport more popular among the population and to demonstrate feasibility. The passengers press a button to their destination and when all passengers are on board the vehicle calculates the shortest route to all chosen destinations. The vehicles communicate with a supervisory system through a radio (University of Washington, 2009). Another upside with automated buses is the costs, approximately 60-70 percent of the total expenditures are salaries to the bus drivers (Johansson, 2004).

5.8 Opinions from Transportforum 2015

An annual Swedish conference, Transportforum, regarding transport was last held in January 2015. Autonomous driving was discussed during the event and experts in the field gave lectures. One of the participants was Neville Stanton, Professor of Human Factors in Transport at the University of Southampton. He discussed that even though self-driving cars are promoted as safe due to reduced human interaction the understanding of the driver might not always be the same as the vehicle’s understanding, which will cause issues. Stanton outs that similarities regarding automation can be seen in previous cases with aircrafts. Plane crashes do happen, even though autopilot controlled aircrafts are seen as uncrashable. Unexpected things will eventually occur while driving an aircraft or another vehicle, which require the driver to take control. Stanton refers to his research about thinking time and movement time of conventional driving compared to the use of self-driving features. The research result shows that a manual driver has a reaction time of 600 milliseconds and the corresponding time for a driver that uses an autonomous vehicle, to take control over the vehicle, is 3.2 seconds. More automation implies more handing over responsibility which gives the driver a new kind of obligation. Humans tend to put little effort in tasks that they do not need to attend to which would increase the thinking and movement times and decrease the road safety in case of an incident. Stanton also raises a question

19 regarding fully automated vehicles that need to be taken into consideration in the introduction phase: “Do we want a relaxed driver?” (Transportforum, 2015).

Other actors who are affected by an introduction of autonomous vehicles participated in the conference. Bengt Svensson from the Swedish police force talked about issues regarding AV’s from the police’s point of view. An occasion he considered as problematic is when the police put up checks or roadblocks along the road by varying reasons. Autonomous vehicles are programmed to avoid obstacles and may not understand that they need to stop. Svensson states that aspects like this are central in the introduction phase and need to be solved before AV’s can enter the commercial market. Another factor that was brought up during the conference was the vehicle’s ability to prioritise actions. There will be times when the vehicle has to prioritise how to handle. For example in a situation where a child runs out into the street the car has to choose to break hard which might hurt both passengers and the child or run into the ditch which also might hurt passengers and damage the car. This means that there will be disagreements regarding how the car will prioritise in such situations. Whether the safety of the driver and its passengers, or the car, is more important than the safety of pedestrians and other entities are other aspects that need to be discussed further before AV’s enter the roads (Transportforum, 2015).

20

6. Simulation Data

In this section the numerical data is presented for both the Vissim simulations with high and low flow and data used for emission calculations in EnViVer.

6.1 Vissim Data

Both maximum and minimum vehicle flows were considered resulting in two simulation groups that both contained all scenarios presented in table 1. Data for the two simulation groups are presented in this section as one unit, because they contain the same simulation parameters except for the vehicle flows. Thus, the simulation presented below is performed twice, one simulation for each flow. All simulations lasted for 5,400 seconds (1.5 hours) but results were collected for time span 1,800-5,400 seconds (1 hour) in order to saturate the system and eliminate the start-up period that begins with an empty system. All changed parameters are presented in the list below.

. Look ahead distance is the minimum and maximum distance that a vehicle can see forward in order to react to other vehicles either in front or to the side within the same link. . Look back distance defines the minimum and maximum distance that a vehicle can see backwards in order to react to other vehicles behind within the same link. . Observed vehicles is a parameter that affects how well vehicles can predict other vehicles’ movements. . Average standstill distance is the average desired distance between two vehicles (see figure 4 and 5). . Smooth closeup behaviour means that vehicles slow down more evenly when approaching a stationary obstacle. . Headway time is the distance in seconds which the driver wants to maintain at a certain speed. The higher the value the more cautious the driver is. . Following variation restricts the distance difference, or how much more distance than the desired safety distance a driver allows before he/she intentionally moves closer to the car in front. . Negative following threshold defines negative speed differences during the following process. A low value results in a more sensitive driver reaction to the acceleration or deceleration of the preceding vehicle. . Positive following threshold defines positive speed differences during the following process. A low value results in a more sensitive driver reaction to the acceleration or deceleration of the preceding vehicle. . Speed dependency of oscillation, the influence on speed oscillation while in following process. If the value is set to zero the speed oscillation is independent of the distance, and if the value is larger than zero it leads to a greater speed oscillation with increasing distance.

21

. Oscillation acceleration is the oscillation during acceleration. . Standstill acceleration is the desired acceleration when starting from standstill. . Acceleration at 80 km/h is the desired acceleration at 80 km/h.

Figure 4. Standstill distance 1.5 meters, corresponds to the average desired distance between two vehicles. Screenshot from Vissim simulation.

Figure 5. Standstill distance 1 meter. It can be seen that the standstill distance is shorter than the one in figure 4. Screenshot from Vissim simulation.

22

Simulation data values were chosen with basis in Bierstedt’s et al. (2014) article considering simulations of an aggressive driving behaviour containing adaptive cruise control. These values are the ones changed in simulations 5-12. Changed simulation parameters are presented in table 1 (Bierstedt et al., 2014).

Table 1. Simulation parameters for the base scenario and the autonomous scenario.

Simulation Parameter Base Scenario Urban Base Scenario Freeway Simulation Scenario

1 Base scenario 2 Look ahead distance Min: 0 m, Max: 150 m Min: 0 m, Max: 250 Min: 0 m, Max: 300/500 m 2 Look back distance Min: 0 m, Max: 50 m Min: 0 m, Max: 150 Min: 0 m, Max: 100/300 m 3 Observed vehicles 4 2 10 4 Smooth closeup behaviour Unchecked Unchecked Checked 5 Standstill distance 1.5 m 2 m 1 m 6 Headway time 0.9 s 1.3 s 0.5 s 7 Following variation 4 m 2 m 1 m 8 Negative following threshold -0.35 -0.35 -0.1 8 Positive following threshold 0.35 0.35 0.1 9 Speed dependency of oscillation 11.44 11.44 0 10 Oscillation acceleration 0.25 m/s2 0.25 m/s2 0.4 m/s2 11 Standstill acceleration 3.5 m/s2 3.5 m/s2 4 m/s2 12 Acceleration at 80 km/h 1.5 m/s2 1.5 m/s2 2 m/s2

13 All parameters

An analysis of the changed parameters was performed before the simulation process started. Simulation number 1 is the base scenario simulation that consists of parameter values typical for Swedish traffic conditions. Next simulation contains two parameter changes, both look ahead distance and look back distance. Autonomous vehicles will be able to gather a lot more data, that can be used for navigation and planning, than a human and these two parameters are therefore hard to estimate with high accuracy. It can be assumed, due to the data collection capacity, that an autonomous vehicle will be able to scan the environment more closely than in the base scenario, and both look ahead distance and look back distance will therefore increase with the introduction. A sensitivity analysis showed that changing the two parameters did not affect the final result strongly. Assumptions and sensitivity analysis weighted resulted in that both distances were doubled compared to the base scenario. Simulation 3 also includes features resulting from the ability to sense the environment in more detail. The number of observed vehicles parameter was therefore set to the maximum amount, 10. Planning ability will also give vehicles the opportunity to drive smoother, which is the feature implemented in simulation number 4.

Parameters for simulations 5-12 are based on Bierstedt’s et al. (2014) article considering simulations of an aggressive driving behaviour containing adaptive cruise control.

23

These parameters have been analysed before implementation in order to determine the accuracy for this study. Standstill distance and following variation are parameters regarding the vehicles ability to navigate with high precision, which will be possible with AV control. It is therefore logical that both parameters are reduced for the autonomous simulations. The speed dependency of oscillation value also depends on the planning ahead ability and is therefore set to zero due to that the speed dependency of oscillation will be eliminated with a 100 percent AV penetration rate. Headway time will be reduced with the introduction of AV’s due to elimination of the human factor. Lind et al. (2014) have performed tests where as short time gaps as 0.2 seconds have been used during platoon driving on a highway at 105 km/h. A headway time of 0.5 seconds was considered as more accurate for the particular simulation network with varying traffic situations. This value corresponds to the one Bierstedt et al. (2014) used for their simulations of an aggressive adaptive cruise control. The following threshold parameters are also connected to a more sensitive and planned journey. Low values result in more sensitive driver reactions to acceleration or deceleration of preceding vehicles.

Simulations 10-12 are all about acceleration which is the parameter that has the most unknown outcome regarding AV’s. Autonomous vehicles will drive smoother and plan the journey to avoid stops but they will also try to avoid congestion and might also drive faster than conventional vehicles. This implies that the acceleration will depend on the traffic situation and sensing technology. The acceleration parameters used for the simulations are therefore assumed to be the same as in Bierstedt’s et al. (2014) simulations, which is slightly higher than values for the base scenario. Temporary lack of attention has two settings; duration and probability. Duration is the period of time when vehicles may not react to a preceding vehicle, and probability is the frequency of the lack of attention. Increased values for these settings will decrease the capacity on the affected links.

6.2 EnViVer Data

Parameters for the emission calculations in EnViVer were set to the same values for all simulations, corresponding to the traffic environment used in previous Vissim simulations. The parameters that can be changed in EnViVer are presented in the list below.

. Road type, can be set to either urban or highway. . Vehicle type: there are three types of vehicles to choose from, light-duty, heavy- duty and bus. . Era represents the year the emissions are calculated for. The default value is 2013 and any year between 2010 and 2025 can be selected. . Fuel type determines the proportion of vehicles that is driven on each fuel type. The fuels that can be selected are petrol, diesel, LPG, CNG and electric. All other fuel types remain zero.

24

. Vehicle age distribution is based on the Swedish vehicle fleet. It is a three- piece linear model and it is set by three parameters; percentage of the vehicles younger than one year, the average vehicle age and the average exit or demolition age. . Emission legislation, the age distribution combined with the specified era will result in a Euro-class distribution. The resulting Euro-class distribution will be updated if changes are made in age distribution, era or introduction date of the Euro-class.

. Average regional CO2 emission: the editor in EnViVer takes the type approval CO2 emission and converts it to real-world CO2 emission.

For the era parameter 2013 was chosen because of available data regarding fuel composition in the Swedish vehicle fleet. The used fuel types were petrol (71%), diesel (28%), CNG (Compressed Natural Gas) (0.9%) and electric (0.1%) (Trafikanalys, 2015). When it comes to vehicle age distribution it was set to values corresponding to the Swedish vehicle fleet in 2013. The average regional CO2 emissions were set to 170 g/km for both petrol and diesel (Trafikverket, 2015).

It is not realistic that an autonomous vehicle fleet with an adoption rate of 100 percent will consist of cars with the same age distribution as today’s fleet. Autonomous vehicles need to be produced forward in time and will therefore most likely have a lower average

CO2 emission rate than 170 g/km, which is used for the calculations. Although, it is necessary to use values for today’s fleet for both CV’s and AV’s to be able to make comparisons between an autonomous fleet and the existing conventional fleet.

25

7. Results

This section presents the results of the study. The results are based on simulations from Vissim and EnViVer. In Vissim three parameters where chosen and analysed; average delay, average number of stops and average speed. This was done for both high vehicle flow and low vehicle flow. As mentioned earlier in the report it is important to keep in mind that it is not representative to look at absolute numerical values. Values depend on user settings and it is therefore more accurate to study differences between results for the conventional fleet and autonomous feature simulations.

7.1 Vissim High Flow

This section first presents figures and tables with absolute numerical values independently for the key indicators and continuously the percentage improvements compared to the base scenario. Numerical results for delay, number of stops and speed are presented in table 2. The high vehicle flow corresponds to a real high flow in the Uppsala traffic network. Even though it is a high traffic load the average speed for the base scenario is 40 km/h. This states that the high flow simulation does not correspond to a scenario with very high congestion where cars are stuck in the network for long periods of time.

Table 2. Numerical resulting values for key indicators for high vehicle flow.

Simulation Changed parameter Delay [s] Number of stops Speed [km/h] 1 Base scenario 173.9 9.43 40.00 2 Look ahead/back distance 125.6 5.84 46.44 3 Observed vehicles 131.2 6.38 45.57 4 Smooth closeup 113.6 4.32 48.27 5 Standstill distance 115.2 4.86 48.10 6 Headway time 113.4 4.32 48.37 7 Following variation 123.4 7.07 46.75 8 Thresholds 124.7 5.52 46.57 9 Speed dependency of oscillation 112.3 5.05 48.60 10 Oscillation acceleration 123.5 5.99 46.75 11 Standstill acceleration 136.3 7.24 44.86 12 Acceleration at 80 km/h 132.6 6.65 45.37 13 All parameters 76.2 3.37 55.37

Delay represents the difference in optimal driving time from start area to the destination. Results from the simulations (see table 2 and figure 6) show that the delay decreases with every parameter change compared to the base scenario. The parameters that make the delay decrease the most are smooth closeup, headway time, speed

26 dependency of oscillation and standstill distance. Speed dependency of oscillation and smooth closeup are both about the vehicle’s ability to plan the driving in terms of other cars. Headway time is connected to reaction time and if the vehicle can act in a shorter amount of time it results in closer time gaps between cars, and therefore a faster journey compared to the one with human reaction time. Change of standstill distance results in shorter delay due to that more vehicles will pass an intersection during green light. The driving is with any of these four parameters changed more optimised and the delay therefore decreases. Highest delay corresponds to the base scenario, and lowest in the scenario with all autonomous vehicle features included.

Delay 200 180 160 140 Average Delay 120 Time [s] 100 80 60 40 20 0

Figure 6. Average delay for high vehicle flow.

Parameters influencing the most when it comes to number of stops are the same parameters as for delay (see figure 7). This result most likely occurs due to the same discussion as above, planning of the driving and shorter time gaps and distances. Number of stops for the base scenario and for the autonomous simulation correspond to the most extreme resulting values which was also the case for the delay.

27

Number of Stops 10 9 8 7 Average Number 6 5 of Stops 4 3 2 1 0

Figure 7. Average number of stops for high vehicle flow.

Similar to previous result parameters, the speed is lowest in the base scenario and highest when all parameters are changed in the autonomous case, which can be seen in figure 8. In this scenario it was desirable to gain a high average speed when driving through the network. Also in this case, the parameters that increased the speed the most were smooth closeup, headway time, speed dependency of oscillation and standstill distance.

Speed

60 50 40 Average Speed [km/h] 30 20 10 0

Figure 8. Average speed for high vehicle flow.

28

In all simulation scenarios the parameter changes influence the driving in a positive way. Delay and number of stops both decrease and speed increases, which imply that vehicles reach their destination faster and more smooth with the introduction of fully automation. Figure 9 presents the percentage improvements for all simulations compared to the base scenario. A resulting decrease of the parameters delay and number of stops corresponds to a percentage improvement and therefore a higher bar in the figure. A higher speed is an improvement in this project and this is as well represented with a higher bar in figure 9. The overall result with all automation parameters included is the one with most improvement in terms of all aspects. Average delay improves with 56 percent, average number of stops with 54 percent and the average speed corresponds to an improvement of 34 percent. Hence, delay and number of stops are more improved than speed in the case with autonomous features included. Delay and number of stops would in an optimal scenario without any obstacles, even though it may not be fully realistic, be cancelled out and become zero. The speed on the other hand has an upper limit, both legally and technologically, that will restrict the improvement of this parameter.

Percentage Improvement Compared to the Base Scenario

60

50

40 Delay Improvement [%] 30 Number of Stops Speed 20

10

0

Figure 9. Percentage improvement of the key indicators for the simulations with all autonomous features changed compared to the base scenario for the high vehicle flow.

29

7.2 Vissim Low Flow

Numerical results for the key indicators delay, number of stops and speed for the low vehicle flow are presented in table 3. Then follows a presentation of the results where autonomous and conventional vehicles are compared and analysed (see figures 10-13).

Table 3. Numerical resulting values for the key indicators during low flow.

Simulation Changed parameter Delay [s] Number of stops Speed [km/h] 1 Base Scenario 44.02 1.04 63.40 2 Look ahead/back distance 45.99 1.11 62.93 3 Observed vehicles 46.42 1.12 62.73 4 Smooth closeup 45.34 1.03 63.13 5 Standstill distance 46.95 1.14 62.62 6 Headway time 45.38 1.08 63.09 7 Following variation 46.18 1.12 62.79 8 Thresholds 45.93 1.10 62.85 9 Speed dependency of oscillation 45.67 1.07 62.85 10 Oscillation acceleration 44.78 1.10 63.15 11 Standstill acceleration 45.63 1.11 62.95 12 Acceleration at 80 km/h 46.74 1.12 62.73 13 All parameters 44.59 1.01 63.16

Results from the simulations show that the key indicators delay, number of stops and speed has a restricted variation for all simulations (see table 3). With low flow the difference between CV’s and AV’s is barely noticeable regarding the selected parameters (see figures 10-12).

Delay 50 45 40 35 Average Delay 30 Time [s] 25 20 15 10 5 0

Figure 10. Average delay for low vehicle flow.

30

Number of Stops 1,4 1,2 1 Average Number 0,8 of Stops 0,6 0,4 0,2 0

Figure 11. Average number of stops for low vehicle flow.

Speed 70 60 50 Average Speed 40 [km/h] 30 20 10 0

Figure 12. Average speed for low vehicle flow.

As for high flow, figure 13 presents the percentage improvements for all simulations compared to the base scenario with low flow. Resulting values for the key indicators are relatively small. Average delay deteriorates with about 1.3 percent, average number of stops improves with 2.9 percent and the average speed corresponds to a deterioration of 0.38 percent. The values are very small and it is therefore not accurate to draw any conclusions if the key indicators will decrease, increase or stay constant for the low vehicle flow.

31

Percentage Improvement Compared to the Base Scenario 10 0 -10 -20 Delay -30 Improvement [%] Number of Stops -40 -50 Speed -60

Figure 13. Percentage improvement of the key indicators for the simulation with all autonomous features compared to the base scenario for low vehicle flow.

7.3 EnViVer Emissions

Results for the emission calculations performed in EnViVer are presented in figure 14 for high vehicle flow. The diagrams show the change of CO2, NOx and PM10 emissions for the base scenario and the simulation with all parameters changed. All three emission categories decreased when implementing autonomous features and the corresponding percentage changes were 20 percent for CO2, 25 percent for NOx and 9 percent for PM10.

Figure 14. CO2, NOx and PM10 emissions for the base scenario and when all parameters are changed for high vehicle flow.

Corresponding results for the low vehicle flow are presented in figure 15. Resulting emissions for the conventional and the autonomous cases simulated with a low flow differs from the presented simulation results for the high flow. All emission types increase slightly and correspond to 0.7 percent for CO2, 3.1 percent for NOx and 1.2

32 percent for PM10 when changing into an automated vehicle fleet. The key indicators delay, number of stops and speed did differ less for the low flow than for the high flow when introducing automation. It is therefore expected that the emission rates also differ marginally for the low flow, which can be seen from the emission calculations.

Figure 15. CO2, NOx and PM10 emissions for the base scenario and when all parameters are changed for low vehicle flow.

7.3.1 Fuel Consumption

To determine fuel consumption calculations were performed using values for average carbon dioxide emissions per litre fuel for the Swedish vehicle fleet. These values are 2.71 kilogram per litre petrol, and 2.89 kilogram per litre diesel (Trafikverket, 2013a). The calculated emissions were converted into litres per kilometre in order to be compared to each other (see table 4 and 5).

Table 4. Fuel consumption for the simulations for high vehicle flow.

High vehicle flow Petrol [l/km] Diesel [l/km] Base scenario 0.0974 0.0913 All parameters 0.0808 0.0758

Table 5. Fuel consumption for the simulations for low vehicle flow.

Low vehicle flow Petrol [l/km] Diesel [l/km] Base scenario 0.0735 0.0690 All parameters 0.0740 0.0694

Fuel consumption decreases or stays constant for both vehicle flows. It can be noticed that the fuel consumption follows the same pattern as previously presented results, it improves more in the high flow case. The calculated fuel consumption for the conventional fleet is relatively high compared to numbers for today’s vehicle fleet. The important aspect to consider in this case too is the percentage changes, which corresponds to a 20 percent decrease for both fuel types for the high flow. Low flow fuel consumption increases with 0.7 percent for both petrol and diesel which matches up with previous findings.

33

8. Discussion

Key indicators studied change with the introduction of autonomous features for both the high and the low flow cases. The changes are much more distinct with a higher number of cars passing through the network and all parameters changed result in an improvement of the key indicators. The low flow reverses almost all results for individual parameters even though the overall result with all parameters changed result in an improvement for one of the parameters and very small decline for the two others. The results from the simulation hence state that an introduction of autonomous vehicles in the Swedish context would have more positive effects on heavy traffic when congestion is reality. This is also reflected in the emission calculations and the fuel consumption. Even though autonomous cars have a higher average speed they emit less than conventional cars. This is due to that AV’s tend to plan their journey and therefore have a smoother ride, which result in less fuel use. Since the results in Vissim were not influenced by the weight of the vehicles this parameter was not changed in the simulations. However, in reality it would matter how much the cars weigh, and lighter vehicles would require less fuel.

A higher traffic flow most likely affects the key indicators in a positive manner due to more effective and planned driving regarding other vehicles. Solitary vehicles do not need to drive close to others or have a full sense of the environment several hundred meters ahead to drive efficiently.

In the simulations made in Vissim some of the characteristics with autonomous vehicles were not possible to implement in the program. In reality the idea is that there will not be necessary to have traffic lights if the penetration rate of AV’s is 100 percent. This is due to the communication between cars, V2V, and communication between cars and the infrastructure, V2I. This feature may increase the efficiency even more compared to the results from the simulations and this will most likely affect both high and low flow in a positive way.

8.1 Adoption and Acceptance

Assumptions made in the simulation model affect the outcome and one of the simplifications that is considered to affect the overall result the most is penetration rate. The rate was set to 100 percent, corresponding to only self-driving cars, in order to obtain clear results. As discussed earlier the time horizon differs for a high adoption rate depending on whom one asks, but it is certain that the composition of the vehicle fleet starts with zero percent autonomous vehicles and goes against a higher rate. This implies that the changes by the introduction initially will be lower than the once carried out from the simulations. The improvement for individual vehicles will also differ from the simulation results due to interaction between the conventional fleet and the autonomous cars. Drivers of CV’s might act differently with the knowledge that some

34 cars might be autonomous and the automated cars need to have larger safety margins, due to lack of needed communication, when interacting in a conventional fleet.

Adoption rate is strongly connected to how people perceive the new technology. Some people might want to drive manually due to different aspects, even though the automated technology is available and has its advantages. Autonomous driving implies that one has to pass on driving tasks, and through that responsibility to safely drive the vehicle, to the control system. Trusting technology might be an issue and will affect penetration rate and resulting effectiveness in the introduction phase.

Another aspect that will get affected by penetration rate is how fast cars will drive. A better safety due to removal of human interaction can lead to that the vehicles can drive faster and still keep passenger’s and other entities’ safety. A small proportion of AV’s would on the other hand result in a scenario where safety margins need to be kept as for the conventional fleet in order to interact with human drivers. This would most likely implicate that the speed of autonomous vehicles will need to be about the same as for conventional vehicles to maintain effectivity and road safety.

Some parameters regarding AV’s might not affect the key indicators studied, but might improve other aspects. Due to lane keeping and parking assistance autonomous vehicles may change current roads and land use. Some researchers believe that automation of cars will enable the vehicles to drive closer to each other, which could also result in more narrow lanes. In the simulations made in this report lateral distance was changed, but a decrease of this parameter did not have any impact on the key indicators, why it is not included in previous sections. It is logical that lateral distance did not have any effect on average delay, average number of stops and average speed. What is likely to be affected by a reduction of the distance between vehicles is urban planning and construction of the roads. Researchers argue that in a case with a low penetration rate AV’s would have to drive in separate lanes, hence a large penetration of AV’s would be required for vast infrastructure changes.

8.2 New Conditions

With automated parking the passenger can be dropped off and the vehicle can locate an available parking space and park close to another vehicle. This would enable smaller parking areas and reduce the need of parking areas in the inner city. This means that highly valued areas could be used for other purposes than parking. Since AV’s would be able to locate a free space for parking, and not have to travel around and search for one, the parking congestion may decrease.

If the driver would not have to pay attention to the driving and could engage in other activities while travelling it would change today’s way of driving, in a case with fully autonomous vehicles and changes of today’s regulations. If the responsibility relies fully on the vehicle it might be possible for people who were previously unable to drive to travel independently by car. For example disabled individuals could in such situations

35 gain significant social benefits, and children could go to school by car on their own. Another possible scenario that comes with law changes is that people will be able to travel individually by car when intoxicated instead of taking a taxi from a party or a bar when the vehicle has all responsibility for driving tasks.

AV’s will make the travel more comfortable, which will reduce the cost per minute and the value of travel time gets less negative. People could for example take a nap, write text messages or answer emails on their way to work, which would affect the value of travel time in a positive way. This is not only positive for each individual; it will also have a positive effect on the whole society if people can be effective on their way to work. And this may lead to that people are more willing to travel farther to their place of work, which in turn leads to that companies can locate their activities outside the city. Companies will benefit from this because it is not as expensive to locate outside the city, and the employees can be effective while travelling. This will in the long term change the urban planning and development in Sweden.

Less negative value of time might make people choose to travel by car instead of public transport. Autonomous vehicles are considered to increase road capacity, which is also shown in the simulation results when considering today’s vehicle composition. A higher number of cars on the roads due to a more positive value of time regarding AV’s compared to public transport might on the other hand increase congestion. The number of occupied road square metres per road user is for a bus 2.1 m2 and for a car the same number is 22.1 m2. This means that it is ten times more efficient to travel by bus than to drive a car in terms of land use. A change to a more car dense vehicle composition together with a higher VMT will affect road capacity but ultimately also land use, emission rates and fuel consumption.

The MTO-theory states that responsibility in case of an accident should be borne equally between man, technology and organization. Conventional driving and current laws put the responsibility on the driver. A shift into full self-driving vehicles, automation level 4, would change this responsibility focus from the driver to the vehicle, from human to technology. It is though not possible to bring technology to justice and this will most likely put the blame on the organisation behind the technology. This states that it is necessary to determine through regulations and laws who is responsible in case of an emergency.

The project has resulted in simulation results together with opinions and discussions about the introduction of autonomous vehicles in the Swedish context. Some of the aspects carried out from the literature review might not be realistic in today’s traffic environment. Aspects such as driving intoxicated and children driving on their own are some of the aspects that are considered as questionable. Both these aspects are illegal today and it will most likely be kept that way for a transparent period of time. The AV technology first has to overcome technical barriers then acceptance and social barriers and finally the regulatory barriers have to be met in order to reach a favourable introduction. It will take a long time to overcome these barriers, which implies that fully

36 autonomous vehicles that are legally responsible for the driving without human interaction will not enter the commercial market in a near future.

8.3 Workshop Discussion

The area of autonomous vehicles is larger than the scope of this report. Some aspects that are not covered of the purpose of this report, but which are still of great value for the discussion of autonomous vehicles, have been enlightened throughout the thesis work. The discussion during the workshop contributed with raising further questions about AV’s effects on society.

One of the aspects discussed is how AV’s might change the way people move. If it gets cheaper and more convenient to travel, more people might be willing to travel by car than today, which would in turn lead to a larger number of private cars. It is almost impossible to determine what effects fully autonomous vehicles will have on travel behaviours, but it is realistic to think that it will change the way people move, which will lead to further changes.

An aspect that questions whether level 4 vehicles are needed in order to gain high safety on the Swedish roads or not was enlightened by the participants. Is it possible to get high safety gains by implementing features that help the driver when needed, such as emergency braking, instead of letting the vehicle have complete control at all times? An introduction of level 4 vehicles will affect all of society and the implementation therefore involves several actors. Conflicts of interest will most likely occur and affect the development of autonomous vehicles. The industry has its incentives to commercialise the technology and society might have others. Some of the conflicts of interest that were discussed during the workshop are presented in table 6. A large part of the discussion contained questions about whom the technology is developed for and which actor is most forcing in the commercialisation.

37

Table 6. Conflicts of interest for the industry and the society.

Industry Society Driving forces Economy User comfort Maintain road traffic Safety Competition, cutting edge Economy Fuel consumption & emissions Travel time Land use Congestion Joint ownership

Opposing forces Laws and regulations Economy Taxes Increased VMT Data safety Data safety Processing large amounts of data Processing large amounts of data User acceptance User acceptance Laws and regulations

Another aspect that was enlightened regarding predicting resulting effects and conflicts of interests was regulations. Some actors want the technology to develop fast while others might want to stop commercialising and further development. Laws and regulations strongly influence the development and can be used either to promote or to prevent it. The regulatory actors therefore stand for an important role in the introduction phase deciding if regulations are changed or not.

8.4 Future Research

Further research is needed in order to both get more detailed simulations and broaden the scope of the implementation in Sweden.

An aspect that needs to be considered in further research is a low penetration rate of autonomous vehicles. Interaction between conventional and autonomous vehicles will affect road capacity and other central aspects on a different scale than a fleet with only CV’s or AV’s. It is therefore important to perform further research with the aim to determine low penetration rate effects that will occur in the introduction phase. Next step is to implement V2V and V2I in the simulations. This would enable the network to function without traffic lights, which would resemble an automated system even more.

Research about further affects that come with the new technology needs to be performed with the aim to determine changes in the long term. What influences the AV’s will have need to be considered fully before regulatory changes that permit use of level 4 automation is reality. The affects are hard to predict before allowing level 4, and it is therefore important to perform further research before commercialising the technology.

38

9. Conclusion

The most critical barriers that need to be considered when introducing autonomous vehicles in the Swedish traffic system are laws and regulations, technology and human and social barriers. The new AV technology will most likely be ready to be used within a few years. The biggest obstacle is today’s laws and regulations, which do not allow what autonomous vehicles require in order to have full responsibility. What is also an important aspect is that AV’s need to be accepted by the society in order to pursue the matter further. It is of importance that people understand that AV’s can make the roads safer and that they can engage in other tasks than driving while travelling, in order to increase the adoption rate and make the implementation favorable.

Simulations show that road capacity, fuel consumption and emission rate of the Swedish car fleet will be improved when introducing autonomous vehicles if the network contains a large proportion of vehicles. It is not accurate to draw any conclusions about changes for the same network containing a low vehicle flow. The pattern with distinct changes for the high vehicle flow and negligible changes for an emptier network can also be seen for emission rate and fuel consumption. Simulation results therefore state that an introduction of autonomous vehicles will improve road capacity, fuel consumption and emission rate in situations with pending congestion.

The workshop contributed with discussions about how autonomous vehicles affect the urban development in the long turn. Three aspects were frequently discussed during the workshop, stating questions for future debate about the introduction. The first aspect is how the introduction will imply behavioral changes and how this will imply further changes in the Swedish traffic system. Second factor is about conflicts of interest between the industry and society, and the third one contains the role of laws and regulations in the introduction phase.

39

References

Anderson, J. M., Nidhi, K., Stanley, K. D., Sorensen, P., Samaras, C., & Oluwatola, O. A. (2014). Autonomous Vehicle Technology: A Guide for Policymakers. Rand Corporation. Anthony, S. (2014). Tesla unveils dual-motor autopilot Model S. New P85D has 691 hp, 0-60 mph in 3.2 s. Available from http://www.extremetech.com/extreme/191811- tesla-unveils-dual-motor-autopilot-model-s-new-p85d-has-691-hp-0-60-mph-in-3-2s [Accessed 13 March 2015] Automotor & Sport, (2013), 100 självkörande Volvos i Göteborg, Available from http://www.automotorsport.se/artiklar/nyheter/20131202/100-sjalvkorande-volvos-i- goteborg [Accessed: 20 February 2015] Barth, M., Boriboonsomsin, K., & Wu, G. (2014). Vehicle Automation and Its Potential Impacts on Energy and Emissions. In Road Vehicle Automation (pp. 103-112). Springer International Publishing. Bierstedt, J., Gooze, A., Gray, C., Peterman, J., Raykin, L., Walters, J., (2014), Effects of Next-Generation Vehicles on Travel Demand and Highway Capacity. Bloomberg Business, (2015), Exclusive: Google Is Developing Its Own Uber Competitor, Available from http://www.bloomberg.com/news/articles/2015-02- 02/exclusive-google-and-uber-are-going-to-war-over-taxis [Accessed: 13 February 2015] Brown, A., Gonder, J., & Repac, B. (2014). An Analysis of Possible Energy Impacts of Automated Vehicle. In Road Vehicle Automation (pp. 137-153). Springer International Publishing. Forward, S. (2008). Driving violations: Investigating forms of irrational rationality. Higgs, B., Abbas, M., & Medina, A. (2011). Analysis of the Wiedemann Car Following Model over Different Speeds using Naturalistic Data. Road Safety and Simulation, Indianapolis, Indiana. Ibañez-Guzman, J., Laugier, C., Yoder, J. D., & Thrun, S. (2012). Autonomous driving: Context and state-of-the-art. In Handbook of Intelligent Vehicles (pp. 1271-1310). Springer London. International Transport Forum, (2014), Autonomous Driving Regulatory Issues, Available from http://www.internationaltransportforum.org/cpb/pdf/autonomous- driving.pdf [Accessed 12 February 2015] Jacobsen, D. I. (2002). Vad och varför? Om metodval I företagsekonomi och andra samhällsvetenskapliga ämnen (1 uppl). Lund: Studentlitteratur. Johansson, T. (2004). Konkurrensegenskaper hos kollektivtrafiksystem baserade på spårvagnar respektive bussar. Väg-och transportforskningsinstitutet.

Knight, W. (2013). Proceed with caution toward the self-driving car. Technology Review, 116(3), 84-86. KPMG, (2012), Self-driving cars: The next revolution KPMG, (2013), Self-Driving Cars: Are we ready? Lind, G., Strömgren, P., Davidsson, F., (2014), Effekter av självstyrande bilar: Litteraturstudie och probleminventering, Movea Kungliga Tekniska Högskolan, (2014), Tekniksprång på väg- Systemeffekter av energieffektiva, autonoma fordon och flottstyrning. Litman, T. (2009). Transportation cost and benefit analysis. Victoria Transport Policy Institute, 1-19. Michael, J. B., Godbole, D. N., Lygeros, J., & Sengupta, R. (1998). Capacity Analysis of Traffic Flow Over a Single-Lane Automated Highway System⋆.Journal of Intelligent Transportation System, 4(1-2), 49-80. National Highway Traffic Safety Administration. (2013). Preliminary statement of policy concerning automated vehicles. Washington, DC. Nåbo, A., Anund, A., Fors, C., & Karlsson, J. G. (2013). Förares tankar om framtida automatiserad bilkörning: en fokusgruppstudie. Pasztor, A. (2014). Tesla Unveils All-Wheel-Drive, Autopilot for Electric Cars. The Wall Street Journal. Available from http://www.wsj.com/articles/tesla-motors-shows- off-automated-driving-system-1412918991 [Accessed 13 March 2015] Peters, J. I. (2014). Accelerating Road Vehicle Automation. In Road Vehicle Automation (pp. 25-35). Springer International Publishing. Pinjari, A. R., Augustin, B., Menon, N., (2013), Highway Capacity Impacts of Autonomous Vehicles: An Assessment. PTV Group, (n.d), What Keeps Traffic Flowing?, Available from http://vision- traffic.ptvgroup.com/fileadmin/files_ptvvision/Downloads_N/0_General/2_Products/2_P TV_Vissim/BRO_PTV_Vissim_EN.pdf [Accessed 5 February 2015] RAND, (2014), Autonomous Vehicle Technology: How to Best Realize its Social Benefits. Available from http://www.rand.org/content/dam/rand/pubs/research_briefs/RB9700/RB9755/RAND_R B9755.pdf [Accessed 28 January 2015] Rigole, P. J. (2014). Study of a Shared Autonomous Vehicles Based Mobility Solution in Stockholm. Sklet, S. (2004). Comparison of some selected methods for accident investigation. Journal of hazardous materials, 111(1), 29-37. TNO. n.d. EnViVer: Model Traffic Flow and Emissions. Available from https://www.tno.nl/en/focus-area/urbanisation/mobility-logistics/clean-mobility/enviver- model-traffic-flow-and-emissions/ [Accessed 30 March]

Trafikanalys, (2015), Fordon i län och kommuner. Statistik 2015:2. Trafikverket, (2013a), Index över nya bilars klimatpåverkan i riket, länen och kommunerna inkl. nyregistrerade kommunägda fordon och dess klimatpåverkan. Trafikverket, (2013b), Kartor med trafikflöden, test point 11830272. Available from http://www.trafikverket.se/Foretag/Trafikera-och-transportera/Trafikera-vag/Verktyg-e- tjanster-och-vagdata/Vagtrafik--och-hastighetsdata/Kartor-med-trafikfloden/ [Accessed: 24 March] Trafikverket, (2015), Fortsatt minskning av utsläppen men i för långsam takt för att nå klimatmålen, H., Johansson. Transportforum, (2015), Inledningen del 2: Autonom körning/autonomous driving (2015). VTI- Transportforum 2015. [video] Available from http://www.vti.se/sv/transportforum/konferensinfo/ [Accessed 23 February 2015] Transportstyrelsen, (2013), Nationell Statistik. Available from http://www.transportstyrelsen.se/sv/vagtrafik/statistik-och- register/Vag/Olycksstatistik/Polisrapporterad-statistik/Nationell-statistik/ [Accessed 29 January 2015] Transportstyrelsen, (2014), Autonom Körning: Förstudie. University of Washington, (2009), Pilot project ParkShuttle, Available from http://faculty.washington.edu/jbs/itrans/parkshut.htm [Accessed: 23 February 2015] Volvo Car Group, (2013), Volvo Car Group initierar världsunikt svenskt pilotprojekt med självkörande bilar på allmän väg, Available from http://www.volvocars.com/se/top/about/news-events/pages/default.aspx?itemid=470 [Accessed 22 January 2015] World Health Organisation, (2013), World report on road traffic injury prevention. Available from http://www.who.int/violence_injury_prevention/publications/road_traffic/world_report/ en/ [Accessed 29 January 2015] Zabat, M., Stabile, N., Farascaroli, S., & Browand, F. (1995). The aerodynamic performance of platoons: A final report. California Partners for Advanced Transit and Highways (PATH).

Workshop participants, 07/05/2015:

Björn Hugosson, consultant in vehicle and traffic strategy, WSP Group

Maria Nilsson, consultant in traffic and energy systems, WSP Group

Patrik Tornberg, consultant in urban planning, WSP Group

Appendix I, Vehicle Flows in Uppsala

An example of the vehicle flow measurements stating the number of vehicles passing by a point in the network, provided by Uppsala Municipality.

Appendix II, Signal Plans & Signal Switching Charts

This section presents the signal plans and switching charts for traffic signals used for both the real network in Uppsala and the simulations in Vissim. Provided by Uppsala Municipality.